PVC pipe injection molding quality optimization method and system based on neural network

Abstract

The invention discloses a method and a system for optimizing the injection molding quality of a pvc pipe fitting based on a neural network, and relates to the technical field of pvc pipe fitting injection molding, the method comprises the following steps: acquiring melt pressure data and screw displacement data in a charging barrel of an injection molding machine, and constructing a filling process curve; performing change point detection analysis according to the filling process curve to generate a first dynamic switching criterion; the switching point of injection and pressure maintaining is adjusted according to the dynamic switching criterion, and injection is switched into pressure maintaining; the pressure maintaining shrinkage rate is predicted according to the filling process curve, and the pressure maintaining pressure is adjusted; according to the adjusted dwell pressure, injection filling and shrinkage compensation are carried out; and the variable point detection analysis further comprises the step of selectively introducing a neural network to assist a regression device to correct a criterion threshold value, and generating a second dynamic switching criterion, so that the problems of bubbles, short shooting and unstable size caused by inaccurate V / P switching opportunity and pressure maintaining shrinkage prediction distortion in the injection molding of the PVC pipe fitting are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of PVC pipe fitting injection molding, more particularly, the present application relates to a PVC pipe fitting injection molding quality optimization method and system based on a neural network. BACKGROUND

[0002] The existing PVC pipe fitting injection molding technology generally includes the following steps: melting PVC particles by a screw injection molding machine, heating and plasticizing, injection molding, pressure maintaining and cooling demolding, etc. During the injection molding process, the control of parameters such as injection pressure, screw speed and cavity pressure is crucial to the molding quality of the pipe fitting. In the traditional process, the operator usually sets the injection speed curve, the pressure maintaining pressure and the V / P switching point based on experience to ensure that the melt can smoothly fill the cavity and to minimize internal defects. In order to cope with PVC pipe fittings of different specifications and wall thicknesses, some high-end injection molding equipment has begun to collect real-time data such as barrel pressure, screw displacement and cavity pressure in order to monitor the process and adjust the parameters, but the overall control strategy is still mainly based on experience.

[0003] Since the traditional V / P switching point and speed curve setting highly depend on experience, they lack real-time response capability to the dynamic characteristics of the PVC pipe fitting melt filling. In actual production, this static setting can easily lead to uneven flow of local melt in the cavity, thereby causing defects such as bubbles, short shots or weld marks. A dynamic switching algorithm based on filling process curves has emerged, which collects pressure, flow and screw displacement data in real time, analyzes the filling process characteristics, and dynamically adjusts the switching time of injection and pressure maintaining in order to reduce the deviation caused by human experience and molding defects.

[0004] However, due to the large volume and slow cooling speed of the PVC pipe fitting, the actual shrinkage rate during the pressure maintaining stage often deviates from the prediction of the static model. This deviation can cause the dynamic switching algorithm based on the filling process curve to distort the model in actual application, the control action is not precise enough, and thus can affect the consistency of pipe fitting molding and the internal quality distribution. In view of the above problems, the present application provides a solution. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a PVC pipe fitting injection molding quality optimization method, system, device and storage medium based on a neural network, which uses a dynamic switching algorithm based on filling process curves and neural network assisted threshold correction to solve the problems of inaccurate V / P switching timing and distorted pressure maintaining shrinkage prediction in PVC pipe fitting injection molding, which can cause bubbles, short shots and size instability.

[0006] To achieve the above object, the present application provides the following technical scheme: A PVC pipe fitting injection molding quality optimization method based on neural network, comprising the following steps: obtaining melt pressure data and screw displacement data in the injection molding machine cylinder, and constructing a filling process curve; performing variable point detection analysis according to the filling process curve to generate a first dynamic switching criterion; adjusting the switching point of injection and pressure maintaining according to the dynamic switching criterion, and switching the injection to pressure maintaining; predicting the pressure maintaining shrinkage rate according to the filling process curve, and adjusting the pressure maintaining pressure; performing injection filling and shrinkage compensation according to the adjusted pressure maintaining pressure; the variable point detection analysis further comprises an optional introduction of a neural network auxiliary regressor for criterion threshold correction to generate a second dynamic switching criterion.

[0007] In a preferred embodiment, the melt pressure data and screw displacement data in the injection molding machine cylinder are obtained, and the filling process curve is constructed, specifically: the screw displacement data includes screw position, screw diameter, screw cross-sectional area, and screw instantaneous speed; the melt pressure data in the injection molding machine cylinder includes filling efficiency, volumetric flow rate, cumulative injection volume, filling percentage, and injection side melt; the melt pressure data and screw displacement data in the injection molding machine cylinder are obtained, time-aligned, and filtered and denoised; the pressure drop caused by the flow channel is estimated based on the power-law fluid empirical pressure drop formula without cavity pressure, and the difference between the estimated pressure drop and the injection side melt pressure is estimated as the cavity pressure in the mold cavity; the volumetric flow rate is calculated based on the screw displacement, and a volumetric flow rate curve varying with time is constructed; the volumetric flow rate curve is integrated with respect to time to obtain the cumulative injection volume; a cumulative injection volume curve, a filling percentage curve, an injection end pressure curve, and a mold cavity pressure curve varying with time are constructed, and a four-dimensional curve family is constructed as the filling process curve.

[0008] In a preferred embodiment, the variable point detection analysis according to the filling process curve generates a first dynamic switching criterion, specifically: the filling process curve is feature extracted to obtain filling process curve features; variable point detection is performed on the filling process curve features based on residual and feature sequence to obtain criterion evidence; the criterion evidence includes pressure rise rate residual cumulative sum pressure rise evidence, flow evidence based on screw speed inflection point, and cavity pressure prediction proximity evidence predicted by short-time mold cavity pressure; the criterion evidence is mapped to a 0-1 confidence score; the criterion evidence mapping includes linear mapping of the pressure rise evidence, proportional mapping of the flow evidence according to a pre-set scaling coefficient, direct assignment of the criterion evidence confidence to 1 when the screw speed inflection point feature meets the set condition, calculation of the proximity based on the difference between the short-time predicted cavity pressure and the real-time measured cavity pressure, and input of the proximity as the confidence; the criterion value is obtained by weighting the confidence score of the criterion evidence, and the first dynamic switching criterion is obtained by setting the judgment logic.

[0009] In a preferred embodiment, the switching point of injection and holding is adjusted according to the dynamic switching criterion, and the injection is switched to holding, specifically: the criterion value at the current time is obtained, and the criterion value is compared with the set upper trigger threshold and lower withdrawal threshold; when the criterion value is greater than or equal to the preset upper trigger threshold and continuously meets the preset first multiple of sampling points, a trigger switching signal is generated; when the criterion value is less than or equal to the preset lower withdrawal threshold and continuously meets the preset second multiple of sampling points, then back off, and a withdrawal switching signal is generated; if the trigger switching signal is valid, a control instruction of injection to holding is immediately generated; if the withdrawal switching signal is valid, the previous switching intention is cancelled, and the injection mode is continued to maintain; the control instruction is sent to the injection molding machine servo drive unit, the execution state of the hydraulic and electric drive module is adjusted, the injection phase is ended and the holding phase is entered; when the maximum injection time or the maximum displacement reaches the mechanical limit, a forced switching protection is triggered.

[0010] In a preferred embodiment, the holding shrinkage rate is predicted according to the filling process curve, and the holding pressure is adjusted, specifically: a first real-time quantity is obtained from the filling process curve, the first real-time quantity includes volume flow rate, filling percentage and holding integral quantity; an initial shrinkage prediction model is established based on a static linear regression model, and after the real shrinkage is obtained in each cycle, the recursive least squares is used for updating to obtain a shrinkage prediction model, and a predicted first shrinkage quantity is output; if the holding phase has not been completed, it is assumed that the injection flow rate in the holding phase decays exponentially, the remaining holding integral quantity is predicted based on exponential decay extrapolation, and the real-time holding integral quantity is added to obtain the completed holding integral quantity; the completed holding integral quantity is used to update the shrinkage prediction model, and a predicted second shrinkage quantity is output, and the difference between the target shrinkage quantity and the predicted second shrinkage quantity is calculated as a shrinkage deviation; the shrinkage deviation is mapped based on linear proportion to a holding pressure adjustment value, and a segmented holding curve is generated; the segmented holding curve is sent to the injection molding machine controller and executed.

[0011] In a preferred embodiment, the shrinkage deviation is mapped based on linear proportion to a holding pressure adjustment value to generate a segmented holding curve, specifically: the shrinkage change caused by the change of holding pressure is determined by experiment to obtain a shrinkage proportion coefficient, which is multiplied by the shrinkage deviation to obtain a recommended holding pressure adjustment amount by linear mapping, a positive value indicating pressure increase and a negative value indicating pressure decrease; the recommended holding pressure adjustment amount is constrained based on safety amplitude limiting and rate limiting, the constraint including single adjustment upper limit constraint, absolute limit and pressure change rate limitation; the holding curve is segmented based on a shape function, the shape function including constant and front-loaded type.

[0012] In a preferred embodiment, the injection filling and shrinkage compensation according to the adjusted holding pressure specifically comprises: performing exponential smoothing on the segmented holding pressure curve, and obtaining a holding pressure adjustment amount in real time; the controller converts the holding pressure adjustment amount into a target valve opening degree recognizable by the injection molding machine, and sends it to the hydraulic controller through the real-time bus; the controller reads the actual pressure and flow of the valve in each sampling period to monitor the tracking error, which is used for feedback and updating of the related prediction model.

[0013] In a preferred embodiment, the variable point detection analysis further comprises optional introduction of a neural network auxiliary regressor for criterion threshold correction to generate a second dynamic switching criterion, specifically: obtaining noise features of the PVC pipe, the noise features including melt temperature fluctuation amplitude, pipe wall thickness ratio and filling non-linear residual; constructing a residual constraint neural network according to the noise features to output a corrected criterion threshold; replacing the threshold in the original set judgment logic with the corrected criterion threshold to obtain the second dynamic switching criterion; the output of the corrected criterion threshold further comprises quality and energy consumption influence prediction of the corrected criterion threshold, which is used to determine whether to correct or return.

[0014] In a preferred embodiment, the quality and energy consumption influence prediction of the corrected criterion threshold, which is used to determine whether to correct or return, specifically comprises: predicting the holding integral amount before and after correction and the shrinkage amount; calculating the absolute difference between the predicted shrinkage amount and the target shrinkage as a quality index; estimating the holding energy consumption difference caused by the change of the trigger point based on the approximate agent method; constructing a cost function as a judgment score based on the quality index and the holding energy consumption difference; determining whether to correct or return according to the judgment score.

[0015] A neural network-based PVC pipe injection quality optimization method and system, comprising a filling process curve acquisition module, a first dynamic switching criterion acquisition module, a holding pressure switching module, a holding pressure adjustment module, an injection molding module, and a switching criterion correction module; the filling process curve acquisition module is used to acquire melt pressure data and screw displacement data in the barrel of the injection molding machine, and construct a filling process curve; the first dynamic switching criterion acquisition module is used to perform variable point detection analysis according to the filling process curve to generate a first dynamic switching criterion; the holding pressure switching module is used to adjust the switching point of injection and holding pressure according to the dynamic switching criterion, and switch the injection to holding pressure; the holding pressure adjustment module is used to predict the holding pressure shrinkage rate according to the filling process curve, and adjust the holding pressure; the injection molding module is used to perform injection filling and shrinkage compensation according to the adjusted holding pressure; and the switching criterion correction module is used to introduce a neural network auxiliary regressor for criterion threshold correction to generate a second dynamic switching criterion.

[0016] The technical effects and advantages of the neural network-based PVC pipe injection quality optimization method and system of the present application are as follows: 1.The present application can analyze the pressure, flow and screw displacement characteristics of PVC pipe fittings during the injection molding filling stage in real time, automatically generate dynamic switching criteria from injection to pressure maintenance, accurately control the V / P switching point and injection speed curve, and effectively avoid defects such as bubbles, short shots or weld marks caused by traditional experience setting, through a dynamic switching algorithm based on filling process curves.

[0017] 2.The present application can dynamically correct the criterion threshold by introducing a small neural network auxiliary regressor, and can dynamically correct the pressure maintenance shrinkage prediction model according to the melt temperature fluctuation, pipe wall thickness ratio and filling nonlinear residual error of the PVC pipe fitting, effectively solve the problem of inconsistent shrinkage rate and model distortion caused by large volume and slow cooling of the PVC pipe fitting, and make the dynamic switching algorithm of the filling process curve accurately applicable in actual production, further improve the shrinkage compensation accuracy and pipe size stability during the pressure maintenance stage. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The present application is a flowchart of a PVC pipe fitting injection molding quality optimization method based on a neural network.

[0019] Figure 2 The present application is a structural diagram of a PVC pipe fitting injection molding quality optimization system based on a neural network. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0021] Embodiment 1, Figure 1 A PVC pipe fitting injection molding quality optimization method based on a neural network is provided, comprising the following steps: S1, obtain melt pressure data and screw displacement data in the barrel of an injection molding machine, and construct a filling process curve.

[0022] In this embodiment, melt pressure data and screw displacement data in the barrel of an injection molding machine are obtained to construct a filling process curve, specifically: The screw displacement data includes screw position, screw diameter, screw cross-sectional area and screw instantaneous speed. The melt pressure data in the injection molding machine barrel includes filling efficiency, volume flow rate, cumulative injection volume, filling percentage, and injection side melt.

[0023] Obtain melt pressure data and screw displacement data in the injection molding machine barrel, and perform time alignment and filtering denoising; Estimate the pressure drop caused by the flow channel based on the power-law fluid empirical pressure drop formula under the condition of no cavity pressure, and estimate the difference between the injection side melt pressure as the cavity pressure in the cavity; Calculate the volume flow rate based on the screw displacement, and construct the volume flow rate curve varying with time; Integrate the volume flow rate curve with respect to time to obtain the cumulative injection volume; Construct the cumulative injection volume curve, the filling percentage curve, the injection end pressure curve, and the cavity pressure curve varying with time, and construct a four-dimensional curve family as the filling process curve.

[0024] It should be noted that the following is a feasible calculation example of the pressure drop caused by the flow channel: ; In the formula, is the pressure drop caused by the flow channel, is the flow channel constant, is the volume flow rate, is the power-law index; It should be noted that the flow channel constant is related to the flow channel length, radius, and material viscosity base, and is determined by a small amount of trial molding and regression. The power-law index is the degree of shear thinning of the material, which is less than 1 for shear thinning, and the typical default value of PVC is in the range of 0.2-1.0, which is obtained by rheological test or trial simulation.

[0025] It should be noted that the following is a feasible calculation example of the volume flow rate: ; In the formula, is the filling efficiency, is the screw cross-sectional area, is the screw speed; It should be noted that the filling efficiency is the ratio of the actual injection amount to the theoretical injection amount.

[0026] The following is a feasible calculation example of the cumulative injection volume: ; In the formula, is the cumulative injection volume, is the injection start time.

[0027] It should be noted that the four-dimensional curve family is that on the time axis t, the four curves share the same horizontal axis, forming a multi-dimensional time sequence curve cluster; the injection side melt pressure is directly obtained by the sensor, and the filling percentage is the ratio of the filling amount to the total capacity.

[0028] It should be noted that the cumulative injection volume, the filling percentage, the injection end pressure and the estimated cavity pressure are placed on the same time axis to form a multi-dimensional curve family, and the change relationship of each curve can be analyzed separately or jointly.

[0029] It should be noted that the original signal collected is filtered and denoised to remove high-frequency noise while retaining the true change trend of pressure and displacement during filling, avoiding fluctuations or misjudgments in subsequent calculations.

[0030] It should be noted that the filling efficiency is calibrated, a correction coefficient is obtained by comparing the actual part volume with the theoretically calculated volume, and is used to correct the injection volume calculated by the screw displacement, so that the volume calculation is closer to the actual injection amount.

[0031] It should be noted that the volume flow rate curve is calculated based on the corrected screw displacement and speed, which reflects the speed and stability of the melt flow in the mold cavity and is an important basis for analyzing the changes in the filling stage.

[0032] It should be noted that in the case of no cavity pressure sensor, the cavity pressure is estimated by an empirical model, the injection end pressure is subtracted from the flow channel pressure drop to obtain an approximate cavity pressure, which is used to monitor the pressure change trend at the end of filling.

[0033] The embodiment accurately acquires screw displacement and barrel pressure data and constructs multi-dimensional filling process curves, so that the short shot, bubbles and weld marks in the filling stage of PVC pipe fittings during injection molding can be identified and quantified in a timely manner, and accurate basic data is provided for the generation of subsequent dynamic switching criteria and pressure holding shrinkage compensation, thereby effectively solving the problems of model distortion and inaccurate control caused by large PVC pipe volume and slow cooling, improving injection molding quality and optimizing energy consumption.

[0034] S2, variable point detection analysis is performed according to the filling process curve to generate a first dynamic switching criterion.

[0035] In the embodiment, variable point detection analysis is performed according to the filling process curve to generate a first dynamic switching criterion, specifically: The filling process curve is characterized to obtain the filling process curve characteristics; The filling process curve characteristics are subjected to variable point detection based on residual and feature sequence to obtain criterion evidence; The criterion evidence includes pressure rise rate residual accumulation and pressure rise evidence, flow evidence based on screw speed inflection point, and cavity pressure prediction proximity evidence predicted by short-time cavity pressure. mapping the criterion evidence into a 0-1 confidence score; The criterion evidence mapping includes linear mapping of pressure rise evidence, proportional mapping of flow rate evidence according to a preset scaling coefficient, directly assigning the criterion evidence confidence as 1 when the screw speed inflection point feature meets a set condition, calculating the closeness according to the difference between the short-time predicted cavity pressure and the real-time measured cavity pressure, and inputting the closeness as the confidence; The criterion value is obtained by weighting the confidence score of the criterion evidence, and a judgment logic is set to obtain the first dynamic switching criterion.

[0036] It should be noted that the following is an example of calculating the cumulative sum of the pressure rise rate residual: ; is the cumulative sum, is the residual of the pressure rise rate, is the allowable deviation constant.

[0037] It should be noted that if is greater than a preset first threshold value, a pressure rise mutation is detected (indicating that the front is blocked or in contact), and the cumulative sum of this evidence is particularly sensitive to small but continuous deviations.

[0038] Linear mapping of pressure rise evidence is specifically mapping the ratio of cumulative sum to trigger threshold value to between 0 and 1.

[0039] It should be noted that the flow rate evidence based on the screw speed inflection point is a sudden drop in flow rate when the screw speed rapidly decreases from a high value and the acceleration is greater than an acceleration threshold value. The following is a condition expression of the feasible candidate sequence number of the flow rate evidence: ; In the formula, is the instantaneous speed of the screw at time t, is the speed decay ratio threshold value, is the maximum instantaneous speed recorded in the current injection cycle, is the screw acceleration, is the acceleration threshold value.

[0040] It should be noted that the mapping of the flow rate evidence is as follows: ; In the formula and are scaling coefficients (used to map the magnitude to a probability perception value), is the mapping value of the flow rate evidence, and if the candidate sequence number strictly meets it, the mapping value is directly 1.

[0041] It should be noted that the short-time cavity pressure prediction cavity pressure prediction proximity evidence is specifically: The ratio of the short-term predicted value based on the polynomial fitting of the cavity pressure based on the recent window to the holding target is used as the cavity pressure prediction proximity evidence.

[0042] It should be noted that the setting judgment logic is specifically: When the criterion value is greater than or equal to the preset upper trigger threshold and continuously satisfies the preset first multiple of sampling points, the injection is switched to holding; When the criterion value is less than or equal to the preset lower threshold and continuously satisfies the preset second multiple of sampling points, it is rolled back; Additionally, when the maximum injection time or maximum displacement reaches the mechanical limit, it is forced to switch.

[0043] It should be noted that the upper trigger threshold is 0.85 by default, the lower threshold is 0.6 by default, the first multiple and the second multiple are 3 by default, and the above can be adjusted according to actual needs.

[0044] It should be noted that the filling process curve feature extraction refers to extracting the change trend and mutation point of the key physical quantity from the four-dimensional curve family, including the pressure change rate, screw displacement speed, injection end pressure and cavity pressure derivative information with respect to time, which is used for subsequent variable point detection criterion calculation; It should be noted that the residual sequence is calculated by the difference between the actual observation value and the predicted value based on the historical injection cycle or the empirical model, which is used to quantify the abnormal deviation phenomenon in the filling process, and can sensitively reflect the situation of PVC melt flow obstruction or retention in large pipes; It should be noted that the pressure rise rate residual accumulation is used to enhance the detection ability of continuous small deviations, which amplifies the small but continuous abnormal trend by accumulating the residual change in a short time, so as to identify the flow front approaching the cavity or the front obstruction in advance; It should be noted that the flow evidence based on the inflection point of the screw speed captures the characteristics of the flow rate drop at the end of the PVC melt filling period by analyzing the inflection point of the instantaneous screw speed and the acceleration threshold condition, which is particularly important in large pipes because the large volume and slow cooling will cause the flow front peak to decay significantly; It should be noted that the short-time cavity pressure prediction proximity evidence is calculated by polynomial fitting and short-term extrapolation prediction of recent cavity pressure data, and the ratio of the predicted value to the holding target is used to quantify the degree of current pressure approaching the target holding state, which is used to determine whether to trigger the holding switch; It should be noted that the criterion evidence mapping is to normalize different physical quantities to the 0-1 interval, so as to unify the weighted calculation of the criterion value, and the selection of linear mapping and scaling coefficient can be customized according to the PVC melt characteristics and pipe volume, to ensure the physical interpretability of the confidence score; It should be noted that the confidence weighting is a step of combining each criterion evidence into a final criterion value according to a preset weight, and the weight can be optimized empirically or data-driven according to the sensitivity of different sizes and cooling conditions of the PVC pipe fitting, to ensure that the criterion value is stable and reliable in response to the switching point; It should be noted that the criterion value judgment logic realizes automatic control of the switching from injection to pressure holding by setting the upper trigger threshold and the lower withdrawal threshold, and the number of consecutive sampling points, in combination with the mechanical limit constraint of the maximum injection time or displacement, to ensure that the PVC pipe fitting can still obtain stable filling and shrinkage compensation under the condition of large volume and slow cooling; It should be noted that the setting of the upper trigger threshold and the lower withdrawal threshold is based on the experimental analysis results of the flow characteristics of PVC melt and the cooling rate, and the threshold selection directly affects the timeliness and accuracy of the pressure holding switching, so it needs to be customized and adjusted under different pipe diameters or wall thicknesses; It should be noted that the continuous sampling point multiple parameter is used to reduce the influence of incidental noise on the switching criterion, and the switching is triggered only by continuously meeting the threshold condition, which improves the robustness of filling abnormalities of large PVC pipe fittings and avoids false switching caused by temporary fluctuations; It should be noted that the residual and feature sequence change point detection method is based on statistical process control and sliding window analysis, which monitors the short-term trend of the pressure rise rate, flow rate and cavity pressure prediction, generates an operable dynamic switching criterion, and realizes fine control in combination with the characteristics of the PVC pipe fitting.

[0045] In this embodiment, the filling process curve is extracted and the change point is detected to generate a dynamic switching criterion, so that the injection molding machine can identify the possible flow front obstruction or front edge in place state of the PVC pipe fitting at the end of the filling period in real time, and ensure the timing accuracy of the switching from injection to pressure holding. This process effectively solves the problem of inaccurate flow front prediction caused by large volume and slow cooling of large PVC pipe fittings in main line 1, and provides a correctable criterion basis for main line 2, so that the subsequent pressure holding pressure adjustment and shrinkage compensation are more reliable, thereby improving the size accuracy and inner and outer surface quality of the molded pipe fitting.

[0046] S3, adjusting the switching point of injection and pressure holding according to the dynamic switching criterion, and switching injection to pressure holding.

[0047] In this embodiment, the switching point of injection and pressure holding is adjusted according to the dynamic switching criterion, and the injection is switched to pressure holding, specifically: Obtain the criterion value at the current time, and compare the criterion value with the set upper trigger threshold and lower withdrawal threshold; When the criterion value is greater than or equal to the preset upper trigger threshold and continuously meets the preset first multiple of sampling points, a switching trigger signal is generated; When the criterion value is less than or equal to the preset lower threshold value and the preset second multiple of sampling points are continuously met, back off, and generate a withdrawal switching signal; If the trigger switching signal is valid, immediately generate a control instruction for injection to pressure holding; If the withdrawal switching signal is valid, cancel the previous switching intention and continue to maintain the injection mode; Send the control instruction to the injection molding machine servo drive unit, adjust the execution state of the hydraulic and electric drive modules, end the injection phase and enter the pressure holding phase; When the maximum injection time or maximum displacement reaches the mechanical limit, trigger the forced switching protection.

[0048] It should be noted that the criterion value is a comprehensive confidence score calculated based on the filling process curve feature extraction and variable point detection, which is used to quantify the change degree of the cavity flow state and pressure rise trend, as the decision basis for injection and pressure holding switching; It should be noted that the upper trigger threshold and the lower withdrawal threshold are used to determine the critical point of switching and back off, respectively. The upper trigger threshold indicates that the cavity has been basically filled and needs to enter the pressure holding phase, and the lower withdrawal threshold is used to avoid early switching to ensure the stability of the filling process and the heat sensitivity compensation of PVC materials; It should be noted that the judgment logic of continuously meeting multiple sampling points is used to filter out the influence of transient disturbance and noise, to ensure the reliability of the switching signal, so that the injection and pressure holding switching responds to the real flow change rather than short-term fluctuations; It should be noted that after the trigger switching signal is generated, the control instruction is sent to the injection molding machine servo drive unit and the hydraulic and electric modules, and by adjusting the screw movement speed and the pressure holding hydraulic pressure, a smooth transition from injection to pressure holding is realized, avoiding short shots, bubbles or weld marks in PVC pipes; It should be noted that the withdrawal switching signal is used to cancel the switching intention when the cavity flow does not reach a stable state, delaying the entry into the pressure holding phase, to prevent the early application of pressure holding from causing uneven PVC shrinkage or surface defects; It should be noted that the mechanical limit protection triggered by the maximum injection time or displacement is used to prevent equipment overload or pipe overfilling, to ensure safety and equipment reliability, while taking into account the cooling characteristics and processing tolerance of PVC large volume pipes.

[0049] The embodiment realizes smooth transition of injection end and pressure maintaining start in the PVC pipe fitting injection molding process by accurately controlling the switching timing of injection and pressure maintaining based on dynamic switching criteria, effectively avoiding defects such as insufficient cavity filling, melt retraction or weld mark caused by switching too early or too late. At the same time, the method ensures the consistency of filling curve characteristics and criterion logic, so that the filling process curve prediction of main line one and the pressure maintaining pressure adjustment strategy can be accurately applied, significantly improving the size stability and surface quality of PVC pipe fittings, while taking into account the dynamic response and material thermal sensitivity compensation problems of the injection molding process.

[0050] S4, adjusting the pressure maintaining pressure according to the pressure maintaining shrinkage rate predicted from the filling process curve.

[0051] In the embodiment, the pressure maintaining pressure is adjusted according to the pressure maintaining shrinkage rate predicted from the filling process curve, specifically: A first real-time quantity is obtained from the filling process curve, the first real-time quantity including volume flow rate, filling percentage and pressure maintaining integral quantity; An initial shrinkage prediction model is established based on a static linear regression model, and after the real shrinkage is obtained in each cycle, the recursive least squares is used for updating to obtain a shrinkage prediction model, and the predicted first shrinkage quantity is output; If the pressure maintaining phase has not been completed, it is assumed that the injection flow rate in the pressure maintaining phase decays exponentially, the remaining pressure maintaining integral quantity is predicted based on exponential decay extrapolation, and the real-time pressure maintaining integral quantity is added to obtain the completed pressure maintaining integral quantity; The completed pressure maintaining integral quantity is used to update the shrinkage prediction model, and the predicted second shrinkage quantity is output, and the difference between the target shrinkage quantity and the predicted second shrinkage quantity is calculated as a shrinkage deviation; The shrinkage deviation is mapped based on linear proportion to a pressure maintaining pressure adjustment value to generate a segmented pressure maintaining curve; The segmented pressure maintaining curve is sent to the injection molding machine controller and executed.

[0052] It should be noted that the pressure maintaining integral quantity is the accumulation result of the injection pressure and the injection flow rate in time after the start of the pressure maintaining phase, which is used to express the total volume effect of the melt being compressed into the cavity under the action of the pressure maintaining pressure, and is a cumulative quantity of the multiplication of pressure, flow rate and time. The following is an example of a feasible calculation: ; In the formula, is the pressure maintaining integral quantity, is the injection pressure at the sampling point k, is the sampling time interval, is the time when the injection is switched to pressure maintaining.

[0053] It should be noted that the following is an example of the model form of the initial shrinkage prediction model established based on the static linear regression model: ; for, is a regression constant term, determined by offline least square method; 、 、 are all influence coefficients, is a holding integral quantity, is a mold temperature, is a cavity volume, is a residual term.

[0054] It should be noted that the shrinkage deviation is mapped based on a linear scale to a holding pressure adjustment value to generate a segmented holding pressure curve, specifically: The shrinkage change caused by a small change in holding pressure is determined by experiment to obtain a shrinkage scale coefficient, which is multiplied by the shrinkage deviation to obtain a recommended holding pressure adjustment amount, a positive value indicating an increase in pressure and a negative value indicating a decrease in pressure.

[0055] The recommended holding pressure adjustment amount is constrained based on a safety limit and a rate limit, including a single adjustment upper limit constraint, an absolute limit, and a pressure change rate limit.

[0056] The holding pressure curve is segmented based on a shape function, which includes a constant and a front-loaded type. The constant type is to add the same pressure to the entire holding segment, and the front-loaded type is to provide more pressure in the early stage and to weaken in the later stage.

[0057] It should be noted that the volumetric flow rate represents the total volume change rate of the melt injected into the cavity per unit time, which is a basic parameter for evaluating filling dynamics and predicting holding shrinkage, and is calculated by screw displacement and screw cross-sectional area.

[0058] It should be noted that the filling percentage is the ratio of the current injected melt volume to the total volume of the cavity, which can directly reflect the degree of cavity filling and is a key reference index for holding shrinkage prediction and holding pressure adjustment.

[0059] It should be noted that the holding integral quantity is the product of the injection pressure and the injection flow rate accumulated with time in the holding stage, which is used to quantify the total volume effect of the melt under the action of holding, and can uniformly express the effects of pressure, flow rate and time, providing a quantitative basis for shrinkage prediction.

[0060] It should be noted that the static linear regression model establishes an initial shrinkage prediction model, which determines the coefficients of each influencing factor, including the holding integral quantity, the mold temperature and the cavity volume, etc., through historical mold testing data, which can quickly estimate the shrinkage trend in the holding stage and provide initial values for subsequent recursive correction.

[0061] It should be noted that the recursive least square update method is used to dynamically correct the model parameters by using the error between the actual shrinkage and the predicted shrinkage after each injection cycle, to realize adaptive prediction of the PVC pipe fitting pressure retention shrinkage, and to improve the prediction accuracy.

[0062] It should be noted that the exponential decay assumption is used to extrapolate the remaining injection flow during the pressure retention stage, which is based on the characteristics of the gradually increasing flow resistance and the gradually decreasing flow rate of the PVC melt during the pressure retention stage. The remaining integral quantity is included in the shrinkage prediction, so that the pressure retention pressure adjustment is more in line with the actual melt behavior.

[0063] It should be noted that the linear proportional mapping is used to convert the shrinkage deviation into the pressure retention pressure adjustment value, and the sensitivity coefficient of shrinkage and pressure retention pressure change is determined by experiment, so that the prediction deviation can be reasonably converted into pressure increase or decrease, and the adjustment amount is controllable and effective.

[0064] It should be noted that the safety limiting and rate limiting are used to constrain the pressure retention pressure adjustment, so as to avoid overpressure of the cavity, melt backflow or pipe fitting warping caused by excessive adjustment amplitude or speed, and to ensure the safety of PVC pipe fitting injection and the quality of finished products.

[0065] It should be noted that the segmented pressure retention curve is designed by using a shape function, and the constant type keeps the same pressure throughout the pressure retention stage, and the front-loaded type has high pressure in the early stage and decreases in the later stage, so that local filling optimization and shrinkage compensation can be realized according to the differences in wall thickness and pipe diameter of PVC pipe fittings.

[0066] In this embodiment, the melt shrinkage and cavity pressure deficiency in PVC pipe fitting injection are actively compensated by predicting and dynamically adjusting the pressure retention pressure based on the filling process curve, which effectively reduces the phenomena of pipe fitting port collapse, uneven wall thickness and warping, and cooperates with the dynamic switching criterion to ensure accurate switching between injection and pressure retention, thereby comprehensively solving the technical problems of insufficient shrinkage compensation accuracy in the filling and pressure retention stages in main line 1 and the adaptability of threshold and switching time in main line 2, and improving the yield and dimensional stability of pipe fittings.

[0067] S5, according to the adjusted pressure retention pressure, injection filling and shrinkage compensation are carried out.

[0068] In this embodiment, according to the adjusted pressure retention pressure, injection filling and shrinkage compensation are carried out, specifically: The segmented pressure retention curve is subjected to exponential smoothing, and the pressure retention pressure adjustment amount is obtained in real time; The controller pressure retention pressure adjustment amount is converted into a target valve opening degree recognizable by the injection molding machine, and is sent to the hydraulic controller through the real-time bus; The controller reads the actual pressure and flow of the valve at each sampling period to monitor the tracking error, which is used for feedback and updating of the related prediction model.

[0069] S6, the variable point detection analysis further comprises an optional introduction of a neural network auxiliary regressor to correct the criterion threshold, to generate a second dynamic switching criterion.

[0070] In this embodiment, S6, the variable point detection analysis further comprises an optional introduction of a neural network auxiliary regressor to correct the criterion threshold, to generate a second dynamic switching criterion, specifically: Obtain the noise characteristics of the PVC pipe fitting, the noise characteristics including melt temperature fluctuation amplitude, pipe fitting wall thickness ratio, and filling non-linear residual error; Construct a residual error constraint neural network according to the noise characteristics, and output a corrected criterion threshold; Replace the threshold value in the original set judgment logic with the corrected criterion threshold value to obtain a second dynamic switching criterion.

[0071] The output corrected criterion threshold further includes quality and energy good impact prediction for the corrected criterion threshold, for determining whether to correct or return.

[0072] It should be noted that the melt temperature fluctuation amplitude is used to reflect that PVC is easily affected by temperature fluctuations, the higher the temperature, the stronger the fluidity, and the switching point is usually advanced.

[0073] It should be noted that different wall thicknesses result in different cooling rates, and the criterion needs to be delayed to trigger, so the pipe fitting wall thickness ratio is used as a training feature; It should be noted that the filling non-linear residual error is obtained by the difference between the injection volume curve and the linear extrapolation model, and is used to describe the hysteresis effect of the PVC flow stage.

[0074] In this embodiment, the quality and energy good impact prediction is performed on the corrected criterion threshold, for determining whether to correct or return, specifically: The pre- and post-pressure holding integral and the shrinkage are predicted; The absolute difference between the predicted shrinkage and the target shrinkage is calculated as a quality index; The pressure holding energy consumption difference caused by the change of the trigger point is estimated based on the approximate proxy method; A cost function is constructed based on the quality index and the pressure holding energy consumption difference as a decision score; Whether to correct or return is determined according to the decision score.

[0075] It should be noted that the following is a calculation example of the feasible decision score: ; In the formula, is the decision score, is the quality index, is the energy consumption weight, is the estimated energy consumption difference, This represents the average energy consumption over one cycle.

[0076] It should be noted that the melt temperature fluctuation amplitude is obtained by placing temperature sensors at different positions in the injection molding machine barrel and cavity, which can accurately reflect the changes in the local flow characteristics of PVC melt and provide a basis for the advance or delay of the judgment threshold. It should be noted that the pipe wall thickness ratio is calculated using mold design parameters and real-time measurement data. It reflects the difference in cooling rate in different areas of the pipe and can be used to correct the switching criteria to avoid uneven wall thickness or collapse caused by switching too early or too late. It should be noted that the nonlinear residual is obtained by calculating the difference between the actual cumulative injection volume curve and the theoretical linear extrapolation model. It can characterize the hysteresis effect and viscoelastic properties of PVC melt in the flow stage, and provide the neural network with nonlinear feature inputs unique to PVC material. It should be noted that during training, the residual constraint neural network combines the above-mentioned PVC features and fits the optimal criterion threshold under different conditions through supervised learning. The corrected threshold is then output for dynamic switching, ensuring that the network prediction results take into account both material properties and filling conditions. It should be noted that the prediction of the impact of quality and energy consumption is achieved by simulating the changes in the pressure holding integral and shrinkage amount triggered by different criterion thresholds. This quantifies the comprehensive impact of criterion adjustments on the quality and energy consumption of the finished product, thereby providing a feasibility assessment and avoiding adverse consequences caused by neural network correction. It should be noted that the judgment score combines the quality index and energy consumption difference, and obtains a comprehensive score through a weighted function. This score is used to determine whether to use the threshold after neural network correction, ensuring that the adjustment can improve the molding accuracy of PVC pipe fittings without significantly increasing energy consumption. It should be noted that the holding pressure integral and shrinkage amount before and after are used to quantify the impact of the change in the criterion threshold on the melt compression effect during the holding pressure stage, and are the basic data for quality assessment and energy consumption calculation. It should be noted that the approximate surrogate method is used to quickly estimate the impact of trigger point changes on pressure holding energy consumption. By reducing the number of repeated simulations, the efficiency of judgment is improved, while maintaining the evaluation accuracy that is sensitive to the injection molding characteristics of PVC pipe fittings.

[0077] This embodiment introduces a residual constraint neural network based on the unique characteristics of PVC pipe fittings to intelligently correct the threshold of the first dynamic switching criterion. It not only considers melt temperature fluctuations, pipe wall thickness differences, and nonlinear hysteresis effects of filling, but also evaluates whether to use a corrected threshold through quality and energy consumption prediction. This effectively compensates for the distortion of the criterion model caused by the susceptibility of PVC material to temperature and flow characteristics in the first main line, achieving more precise control of the injection-holding pressure switching point, improving the consistency and quality of pipe fitting molding, and avoiding increased energy consumption caused by blind adjustments.

[0078] Embodiment 2, Figure 2 The application discloses a PVC pipe fitting injection quality optimization system based on a neural network, which comprises a filling process curve acquisition module, a first dynamic switching criterion acquisition module, a pressure maintaining switching module, a pressure maintaining pressure adjustment module, an injection molding module and a switching criterion correction module. The filling process curve acquisition module is used for acquiring melt pressure data and screw displacement data in a barrel of an injection molding machine and constructing a filling process curve. The first dynamic switching criterion acquisition module is used for performing variable point detection analysis according to the filling process curve and generating a first dynamic switching criterion. The pressure maintaining switching module is used for adjusting a switching point of injection and pressure maintaining according to the dynamic switching criterion and switching the injection to pressure maintaining. The pressure maintaining pressure adjustment module is used for predicting a pressure maintaining shrinkage rate according to the filling process curve and adjusting the pressure maintaining pressure. The injection molding module is used for performing injection filling and shrinkage compensation according to the adjusted pressure maintaining pressure. The switching criterion correction module is used for introducing a neural network auxiliary regressor to correct a criterion threshold value and generate a second dynamic switching criterion.

[0079] In the embodiments of the present application, it should be understood that the disclosed system can be implemented in other manners. For example, the system embodiments described above are merely schematic; for example, the division of the modules is only a logical function division; and there can be another division manner in actual implementation.

[0080] In addition, each function module in each embodiment of the present application can be integrated in a processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware, or in the form of hardware plus software function modules.

[0081] Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be regarded as limiting the claims to which they relate.

[0082] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0083] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0084] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of functional units and modules is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the system is divided into different functional units or modules to complete all or part of the functions described above.

[0085] In the embodiments provided in the present disclosure, it should be understood that the disclosed system and method can also be implemented in other manners. The above described system embodiments are merely illustrative, for example, the flowcharts and block diagrams in the accompanying drawings show possible implementation architectures, functions and operation of the system, method and computer program product according to the embodiments of the present disclosure. In this regard, each block in the flowcharts or block diagrams can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in a different order than that noted in the accompanying drawings. For example, two consecutive blocks can actually be executed substantially in parallel, or they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and the combination of blocks in the block diagrams and / or flowcharts, can be implemented by a dedicated hardware-based system, or by a combination of special-purpose hardware and computer instructions.

[0086] It should be noted that in the present disclosure, the term "comprising" or "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent in such a process, method, article or apparatus. Without more limitations, the element limited by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0087] The above described embodiments are merely used to illustrate the technical solutions of the present application, rather than limiting them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A neural network-based PVC pipe fitting injection quality optimization method, characterized in that, The method comprises the following steps: obtaining melt pressure data and screw displacement data in the injection molding machine barrel, and constructing a filling process curve; detecting a variable point based on the filling process curve to generate a first dynamic switching criterion; adjusting the switching point of injection and pressure maintaining based on the dynamic switching criterion, and switching the injection to pressure maintaining; adjusting the pressure maintaining pressure based on the shrinkage rate predicted by the filling process curve; adjusting the pressure maintaining pressure based on the shrinkage rate predicted by the filling process curve; The variable point detection analysis further comprises an optional introduction of a neural network auxiliary regressor for threshold correction of the criterion, to generate a second dynamic switching criterion.

2. The neural network-based PVC pipe fitting injection quality optimization method according to claim 1, wherein, The melt pressure data and screw displacement data in the injection molding machine barrel are obtained, and the filling process curve is constructed, specifically as follows: The screw displacement data comprises screw position, screw diameter, screw cross-sectional area and screw instantaneous speed; The melt pressure data in the injection molding machine barrel comprises filling efficiency, volumetric flow rate, cumulative injection volume and filling percentage; The melt pressure data and screw displacement data in the injection molding machine barrel are obtained, and time alignment and filtering denoising are performed; The pressure drop caused by the flow channel is estimated based on the power-law fluid empirical pressure drop formula under the condition of no cavity pressure, and the difference between the estimated pressure drop and the melt pressure on the injection side is estimated as the cavity pressure in the cavity; The volumetric flow rate is calculated based on the screw displacement, and a volumetric flow rate curve varying with time is constructed; The cumulative injection volume is obtained by time integration of the volumetric flow rate curve; The cumulative injection volume curve, the filling percentage curve, the injection end pressure curve and the cavity pressure curve varying with time are constructed, and a four-dimensional curve family is constructed as the filling process curve.

3. The neural network-based PVC pipe fitting injection quality optimization method of claim 2, wherein, The variable point detection analysis based on the filling process curve generates a first dynamic switching criterion, specifically as follows: feature extraction is performed on the filling process curve to obtain filling process curve features; variable point detection is performed on the filling process curve features based on residual and feature sequence to obtain criterion evidence; The criterion evidence comprises pressure rise rate residual cumulative sum pressure rise evidence, flow rate evidence based on screw speed inflection point, and cavity pressure prediction proximity evidence predicted by short-time cavity pressure; The criterion evidence is mapped to a 0-1 confidence score; The criterion evidence mapping comprises linear mapping of the pressure rise evidence, proportional mapping of the flow rate evidence according to a preset scaling coefficient, direct assignment of the criterion evidence confidence to 1 when the screw speed inflection point feature meets the set condition, calculation of the proximity based on the difference between the short-time predicted cavity pressure and the real-time measured cavity pressure, and input of the proximity as the confidence; The criterion value is obtained by weighting the confidence score of the criterion evidence, and the judgment logic is set to obtain the first dynamic switching criterion.

4. The neural network-based PVC pipe fitting injection quality optimization method of claim 3, wherein, The switching point of injection and pressure maintaining is adjusted based on the dynamic switching criterion, and the injection is switched to pressure maintaining, specifically as follows: The criterion value at the current time is obtained, and the criterion value is compared with the set upper trigger threshold and lower withdrawal threshold; When the criterion value is greater than or equal to the preset upper trigger threshold and continuously meets the preset first multiple sampling points, a trigger switching signal is generated; When the criterion value is less than or equal to the preset lower withdrawal threshold and continuously meets the preset second multiple sampling points, a withdrawal switching signal is generated. If the switch signal is valid, the control instruction of injection and pressure maintaining is generated immediately; If the switch signal is valid, the previous switch intention is cancelled, and the injection mode is maintained continuously; The control instruction is sent to the servo drive unit of the injection molding machine to adjust the execution state of the hydraulic and electric drive modules, end the injection phase and enter the pressure maintaining phase; When the maximum injection time or the maximum displacement reaches the mechanical limit, the forced switch protection is triggered.

5. The neural network-based pvc pipe fitting injection quality optimization method according to claim 4, wherein, The shrinkage rate during pressure maintaining is predicted according to the filling process curve, and the pressure maintaining pressure is adjusted, specifically as follows: A first real-time quantity is obtained from the filling process curve, and the first real-time quantity includes a volume flow rate, a filling percentage and a pressure maintaining integral quantity; An initial shrinkage prediction model is established based on a static linear regression model, and after the real shrinkage is obtained in each cycle, the shrinkage prediction model is updated by using a recursive least square to obtain a predicted first shrinkage quantity; If the pressure maintaining phase has not been completed, it is assumed that the injection flow rate during the pressure maintaining phase decays exponentially, the remaining pressure maintaining integral quantity is predicted based on the exponential decay extrapolation, and the real-time pressure maintaining integral quantity is added to obtain a completed pressure maintaining integral quantity; The completed pressure maintaining integral quantity is used to update the shrinkage prediction model, and a predicted second shrinkage quantity is output, and a difference between the predicted second shrinkage quantity and a target shrinkage quantity is calculated as a shrinkage deviation; The shrinkage deviation is mapped to a pressure maintaining pressure adjustment value based on a linear proportion, and a segmented pressure maintaining curve is generated; The segmented pressure maintaining curve is sent to the injection molding machine controller and executed.

6. The neural network-based PVC pipe fitting injection quality optimization method of claim 5, wherein, The shrinkage deviation is mapped to the pressure maintaining pressure adjustment value based on the linear proportion, and the segmented pressure maintaining curve is generated, specifically as follows: A shrinkage proportion coefficient is obtained by testing the shrinkage change caused by the change of the pressure maintaining pressure, and the shrinkage proportion coefficient is multiplied by the shrinkage deviation to obtain a recommended pressure maintaining pressure adjustment quantity, and a positive value indicates pressure increasing and a negative value indicates pressure decreasing; The recommended pressure maintaining pressure adjustment quantity is constrained based on a safety limit and a rate limit, and the constraint includes a single adjustment upper limit constraint, an absolute limit and a pressure change rate limit; The pressure maintaining curve is segmented based on a shape function, and the shape function includes a constant and a front-loaded type.

7. The neural network-based PVC pipe fitting injection quality optimization method of claim 6, wherein, The injection filling and shrinkage compensation are performed according to the adjusted pressure maintaining pressure, specifically as follows: The segmented pressure maintaining curve is subjected to exponential smoothing processing, and a pressure maintaining pressure adjustment quantity is obtained in real time; The controller pressure maintaining pressure adjustment quantity is converted into a target valve opening degree recognizable by the injection molding machine, and is sent to the hydraulic controller through a real-time bus; The controller reads the actual pressure and flow of the valve in each sampling cycle to monitor the tracking error, which is used for feedback updating of the related prediction model.

8. The neural network-based PVC pipe fitting injection quality optimization method of claim 7, wherein, The variable point detection analysis further includes an optional introduction of a neural network auxiliary regressor to correct the criterion threshold to generate a second dynamic switching criterion, specifically as follows: Noise characteristics of the PVC pipe fitting are obtained, and the noise characteristics include a melt temperature fluctuation amplitude, a pipe fitting wall thickness ratio and a filling nonlinear residual error; A residual error constraint neural network is constructed according to the noise characteristics to output a corrected criterion threshold; The corrected criterion threshold is used to replace the threshold in the original set judgment logic to obtain the second dynamic switching criterion; The output of the corrected criterion threshold further includes a quality and energy influence prediction of the corrected criterion threshold, which is used to determine whether to correct or return.

9. The neural network-based pvc pipe fitting injection quality optimization method of claim 8, wherein, The criterion threshold value is corrected to predict the influence of quality and energy consumption, and whether to correct or return is judged, specifically: The pre-filling and post-filling pressure integral quantities and shrinkage are predicted; The absolute difference between the predicted shrinkage and the target shrinkage is calculated as a quality index; The pressure-holding energy consumption difference caused by the change of the trigger point is estimated based on the approximate agent method; A cost function is constructed based on the quality index and the pressure-holding energy consumption difference as a decision score; Whether to correct or return is determined according to the decision score.

10. A system for using the neural network-based pvc pipe fitting injection molding quality optimization method according to any one of claims 1-9, characterized in that, It comprises a filling process curve acquisition module, a first dynamic switching criterion acquisition module, a pressure-holding switching module, a pressure-holding pressure adjustment module, an injection molding module, and a switching criterion correction module. The filling process curve acquisition module is used to acquire the melt pressure data and screw displacement data in the injection molding machine cylinder, and to construct the filling process curve. The first dynamic switching criterion acquisition module is used to perform variable point detection analysis based on the filling process curve to generate the first dynamic switching criterion. The pressure-holding switching module is used to adjust the switching point of injection and pressure-holding according to the dynamic switching criterion, and to switch injection to pressure-holding. The pressure-holding pressure adjustment module is used to predict the pressure-holding shrinkage rate based on the filling process curve, and to adjust the pressure-holding pressure. The injection molding module is used to perform injection filling and shrinkage compensation based on the adjusted pressure-holding pressure. The switching criterion correction module is used to introduce a neural network auxiliary regressor to correct the criterion threshold value to generate the second dynamic switching criterion.

Images